Pith. sign in

REVIEW 1 major objections 7 minor 50 references

Iterative clustering and imputation co-refine each other

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · glm-5.2

2026-07-09 22:43 UTC pith:25PDQ2MJ

load-bearing objection Solid empirical paper on cluster-conditioned GAN imputation; the core mechanism claim is under-validated but the method works. the 1 major comments →

arxiv 2607.06930 v1 pith:25PDQ2MJ submitted 2026-07-08 cs.LG cs.AI

Imputation Meets Clustering: Exploiting Latent Subgroup Structure for Missing Data Recovery

Chuyao Zhang , E Li , Taochen Chen , Yiqun Zhang , Yuzhu Ji , Shuping Zhao , Peng Liu , Yiu-ming Cheung This is my paper
classification cs.LG cs.AI
keywords cagiimputationsubgroupdataclusterclusteringdatasetsdistributions
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper argues that missing data imputation and subgroup discovery are two halves of a single problem that should be solved together rather than sequentially. In real datasets containing multiple subgroups with distinct statistical profiles, a single global imputation model produces values that average across subgroups, landing in regions plausible for no particular group. The proposed method, CAGI, breaks this by first assigning incomplete samples to clusters using a partial-distance metric that works even when features are missing, then conditioning a generative adversarial network on those cluster labels so it imputes values consistent with the local subgroup distribution. The completed data is periodically re-clustered, and the refined cluster assignments feed back into the generator. This alternating loop resolves the circular dependency between needing complete data to cluster and needing cluster labels to impute. A multi-level objective anchors the process: reconstruction loss on observed entries provides ground-truth supervision, adversarial training enforces element-level realism, and an optimal-transport regularizer prevents the cluster-conditioned generator from fragmenting the global distribution. Experiments on 14 datasets against 15 baselines show the approach yields lower imputation error, particularly when subgroup structure is pronounced.

Core claim

The central discovery is that clustering and imputation can be made mutually reinforcing through an alternating optimization loop where cluster assignments serve as conditioning signals for a generative imputation model, and the resulting imputations are periodically re-clustered. This co-optimization outperforms both global imputation methods that ignore subgroup structure and two-stage cluster-then-impute pipelines that fix cluster assignments prematurely.

What carries the argument

The key machinery is a three-part loop: (1) a missing-tolerant clustering step using partial Euclidean distances computed only over co-observed features, scaled to approximate full-dimensional distance; (2) a GAN-based generator conditioned on cluster assignments that imputes missing entries, combined with observed values via a mask; and (3) a composite loss combining adversarial discrimination, reconstruction on observed entries, and Sinkhorn-divergence optimal-transport regularization between random partitions of completed mini-batches. Every T iterations, the generator's output is re-clustered and the new assignments replace the old, creating an imputation-reclustering cycle.

Load-bearing premise

The method assumes that the initial partial-distance clustering on incomplete data produces assignments sufficiently aligned with true latent subgroups to bootstrap the feedback loop. If the initial clusters are badly misaligned, the loop could reinforce its own errors, converging to a local optimum where wrong clusters and wrong imputations stabilize each other.

What would settle it

A dataset where the true subgroup structure is orthogonal to the feature space available for partial-distance computation, so initial clustering is essentially random; if the feedback loop still converges to correct subgroup structure, the self-correction claim is strong, but if it locks into wrong clusters and produces worse imputation than a global model, the approach fails.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Datasets with strong latent subgroup structure, such as clinical diagnostics with disease subtypes or marketing data with demographic segments, would benefit most from cluster-conditioned imputation rather than monolithic global models.
  • The partial-distance clustering metric provides a general mechanism for bootstrapping any iterative algorithm that requires labels on incomplete data, beyond imputation to tasks like semi-supervised learning with missing features.
  • The finding that the number of clusters K need not match the true subgroup count suggests the method is robust to over-partitioning, which matters for practical deployment where true subgroup counts are unknown.
  • The optimal-transport regularization preventing distributional fragmentation could generalize to other conditional generation settings where conditioning risks isolating sub-distributions from the global manifold.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

1 major / 7 minor

Summary. The paper proposes CAGI (Cluster-Aware Generative Imputation), a framework that jointly optimizes clustering and imputation via a 'Partition-Guide-Restore' strategy. The core idea is to use dynamic cluster assignments as conditioning signals for a GAN-based generator, with periodic reclustering on imputed data to iteratively refine both subgroup structure and imputed values. A multi-level objective combines instance-level reconstruction (L_MSE), adversarial training (L_ADV), and distribution-level optimal transport regularization (L_OT via Sinkhorn divergence). Experiments span 14 datasets (numerical, categorical, mixed-type) and 15 baselines, with ablation studies isolating each component. The central claim is that decomposing the global distribution into subgroup-conditional submanifolds yields higher-fidelity imputations than monolithic approaches.

Significance. The paper addresses a well-motivated problem: the tension between subgroup heterogeneity and global imputation. The co-optimization design, where cluster assignments are treated as tunable latent variables rather than fixed preprocessing outputs, is a reasonable architectural contribution. The experimental scope is commendable—14 datasets, 15 baselines, 5-fold cross-validation with 5 repetitions, and ablation isolating each loss component and design choice. The release of source code is a positive for reproducibility. The sensitivity analysis over K (number of clusters) showing robustness across a wide range is a practically useful finding.

major comments (1)
  1. Section 1, Figure 1: The paper's central conceptual claim is that existing methods fail because they conflate multiple subgroups into a global average, and CAGI's gains come from recovering subgroup-conditional distributions. However, the 14 benchmark datasets (e.g., spam, breast, wine, adult) are standard UCI datasets whose suitability for testing latent subgroup heterogeneity is not established. The paper does not demonstrate that these datasets actually contain the distinct subgroups the method is designed to exploit. If the datasets are approximately unimodal, CAGI's gains cannot be attributed to subgroup recovery—they could stem from the regularization effects of the multi-level loss or cluster-conditioning acting as a generic mixture-of-experts regularizer. The ablation (Table 3) confirms that removing clustering degrades performance, but this does not confirm the degradation is *w
minor comments (7)
  1. Section 1: The paper states it focuses on MCAR, excluding MAR and MNAR. This is a reasonable scoping decision but limits the practical applicability claims. A brief discussion of whether the co-optimization framework could extend to MAR/MNAR would strengthen the paper.
  2. Table 1: On the 'blood' dataset, CAGI (RMSE=.1539) is outperformed by MissForest (.1270), MIWAE (.1349), GRAPE (.1325), and DiffPuter (.1421). The paper should acknowledge datasets where CAGI is not the best method and discuss conditions under which the cluster-conditioning approach is less effective.
  3. Section 3.2, Eq. (7): The noise initialization uses a variance factor of 0.1. While the paper states this is not tightly coupled to final performance, a brief justification or sensitivity analysis for this specific choice would be helpful.
  4. Figure 3: The stacked bar chart showing RMSE and PFC simultaneously is difficult to read. A table format or separate panels would improve clarity for the mixed-type results.
  5. Section 3.3, Eq. (13): The Sinkhorn divergence is computed on mini-batch-sized subsets. The reliability of OT estimates on small batches is a known concern; a brief comment on batch size and its interaction with epsilon would be useful.
  6. The paper lists 8 free hyperparameters (K, alpha, beta, gamma, p_h, epsilon, T, noise variance). While sensitivity analysis is provided for some, a table of all hyperparameter values used across experiments would improve reproducibility.
  7. Section 2.2: The related work on unsupervised pattern analysis is brief. Given that clustering is a core component, a more thorough discussion of how CAGI's clustering module relates to recent mixed-data clustering advances would be an opportunity for improvement.

Circularity Check

0 steps flagged

No circularity found; the iterative loop is an alternating optimization algorithm, not a definitional equivalence, and all load-bearing components are externally sourced.

full rationale

The paper's central contribution is an iterative imputation-clustering loop (Section 3.4, Eqs. 15-16) that alternates between generating imputed data and re-clustering. While the paper explicitly frames this as resolving a 'circular dependency' (Section 1: 'reliable subgroup identification requires complete data, while complete data is the very objective of imputation'), this is a problem statement, not a logical circularity in the derivation. The resolution is a standard alternating optimization procedure with empirical convergence, evaluated against 15 external baselines on 14 public datasets using standard metrics (RMSE, PFC, AUROC, ARI). All load-bearing components are built on independently published work: GAIN [33, Yoon et al.], Sinkhorn divergence [8, Cuturi], K-Prototypes [12, Huang], and partial-distance clustering [36, 23]. The self-citations by Zhang/Cheung ([37-41, 45, 46]) appear only in the Related Work section (Section 2.2) as examples of clustering algorithm progress and are not invoked as load-bearing justification for any design choice or claim in CAGI. No 'prediction' is derived from a fitted parameter and presented as a first-principles result. No uniqueness theorem is invoked. The derivation chain is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

8 free parameters · 4 axioms · 0 invented entities

The paper introduces no new mathematical entities, physical particles, or theoretical constructs. It combines existing algorithms (K-Means/K-Prototypes, GAN, Sinkhorn divergence) into a new architecture. The free parameters are standard hyperparameters for GAN and clustering methods. The axioms are domain assumptions about data structure and estimator reliability, plus one unstated convergence assumption.

free parameters (8)
  • K (number of clusters) = 5 (default)
    Number of clusters for K-Means/K-Prototypes; paper shows robustness across K values but K is still a user-specified parameter.
  • alpha (L_MSE weight) = tuned via grid search in [0.5, 500]
    Controls reconstruction loss weight in the composite objective (Eq. 14).
  • beta (L_OT weight) = tuned via grid search in [0.1, 5]
    Controls Sinkhorn regularization weight in the composite objective (Eq. 14).
  • gamma (categorical distance weight) = not specified
    Balances numerical and categorical dissimilarity in K-Prototypes distance (Eq. 5); value not reported.
  • p_h (hint rate) = not specified
    Bernoulli parameter for hint matrix generation (Eq. 9); value not reported in paper.
  • epsilon (Sinkhorn regularization) = not specified
    Entropy regularization parameter for Sinkhorn divergence (Eq. 13); value not reported.
  • T (cluster update frequency) = 500 (default)
    Number of iterations between cluster re-assignments; sensitivity analyzed in Figure 6b.
  • noise variance factor = 0.1
    Scales the standard deviation for initial noise filling of missing entries (Section 3.2).
axioms (4)
  • domain assumption Real-world datasets contain latent subgroup structure with distinct distributions that can be recovered by clustering.
    Foundational premise of the paper; stated in Section 1 and illustrated in Figure 1. Not proven but reasonable for many practical datasets.
  • domain assumption Partial distances computed on co-observed features provide a sufficient approximation of full-dimensional distances for initial clustering.
    Invoked in Section 3.2 (Eq. 4) to enable K-Means on incomplete data. The scaling factor d/|S_ij| assumes features are equally informative, which may not hold.
  • domain assumption The Sinkhorn divergence between random mini-batch partitions provides a reliable estimate of distributional self-consistency.
    Invoked in Section 3.3 (Eq. 13). The paper cites Cuturi 2013 for the metric but the specific use for self-consistency regularization on mini-batches is an unstated assumption about estimator reliability.
  • ad hoc to paper The alternating imputation-clustering loop converges to a stable equilibrium.
    The paper claims convergence in Figure 4 caption ('stable equilibrium') but provides no formal proof or convergence analysis.

pith-pipeline@v1.1.0-glm · 17101 in / 2799 out tokens · 295884 ms · 2026-07-09T22:43:43.462959+00:00 · methodology

0 comments
read the original abstract

Missing data is prevalent in practical applications, making effective imputation an essential preprocessing step for downstream analysis. Real-world datasets often exhibit complex latent structures composed of multiple subgroups with distinct distributions. However, existing methods often overlook such population heterogeneity. Without explicit structural guidance, these methods tend to produce generic estimates that blur subgroup boundaries and lack instance-level fidelity. While incorporating subgroup information offers a remedy, it faces a circular dependency: reliable subgroup identification requires complete data, while data completion is the imputation objective itself. To resolve this, we propose CAGI (Cluster-Aware Generative Imputation), a framework that reformulates clustering and imputation as a mutually reinforcing co-optimization process. CAGI employs a ``Partition-Guide-Restore'' strategy where dynamic cluster assignments act as local priors to condition a Generative Adversarial Network. An iterative feedback loop is established to progressively refine both cluster structures and imputed values toward faithful subgroup distributions. To ensure distributional stability, CAGI further employs a multi-level optimization objective combining instance-level reconstruction with distribution-level regularization. Extensive experiments on 14 benchmark datasets with 15 representative baselines demonstrate the superiority of CAGI. The source code is available at: https://github.com/supercocachii/CAGI

Figures

Figures reproduced from arXiv: 2607.06930 by Chuyao Zhang, E Li, Peng Liu, Shuping Zhao, Taochen Chen, Yiqun Zhang, Yiu-ming Cheung, Yuzhu Ji.

Figure 1
Figure 1. Figure 1: Motivating example of imputation under subgroup heterogeneity in an unsu￾pervised setting. Subgroup colors are shown only for illustration. (a) Incomplete data with two latent subgroups whose memberships are unknown. (b) Global imputation, agnostic to latent structure, deviates toward the global average and produces misplaced values. (c) CAGI discovers subgroup structure (shaded regions indicate learned cl… view at source ↗
Figure 2
Figure 2. Figure 2: The overall architecture of CAGI framework following a “Partition-Guide￾Restore” design: the missing-tolerant clustering module first assigns the incomplete data into clusters and treats the labels as conditioning signals for generator G. Then G imputes all positions to produce x¯, and the completed data xˆ retains original observed values and replaces missing entries with x¯ mis ij , which is fed to discr… view at source ↗
Figure 3
Figure 3. Figure 3: Imputation performance comparison on mixed-type datasets. Each stacked bar shows RMSE (top) and PFC (bottom) simultaneously [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: The contribution ra￾tio of each loss component (LMSE, LADV, and LOT) [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: Sensitivity to: (a) number of clusters K, (b) cluster update frequency, and (c) Sinkhorn computation frequency. Default setting: K = 5, cluster update frequency= 500, Sinkhorn frequency= 5. 4.4 Sensitivity Analysis This subsection examines the sensitivity of CAGI to key hyperparameters (i.e., Loss Weights, Number of Clusters, and Update Frequencies). Loss Weights [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Classification performance (top, measured by AUROC) and clustering perfor￾mance (bottom, measured by ARI) on spam, letter, and adult. Higher values indicate better performance for both metrics. 4.5 Downstream Task Evaluation This subsection evaluates whether CAGI’s improved imputation fidelity trans￾lates into tangible gains for classification and clustering. Classification Performance. As shown in [PITH_… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages · 7 internal anchors

  1. [1]

    Journal of Computer and Communications12, 53–75 (2024)

    Alwateer, M., Atlam, E.S., Abd El-Raouf, M.M., et al.: Missing data imputation: A comprehensive review. Journal of Computer and Communications12, 53–75 (2024)

    work page 2024

  2. [2]

    In: Proceedings of the 34th International Conference on Machine Learning

    Arjovsky,M.,Chintala,S.,Bottou,L.:Wassersteingenerativeadversarialnetworks. In: Proceedings of the 34th International Conference on Machine Learning. pp. 214–223. PMLR (2017)

    work page 2017

  3. [3]

    Health and quality of life outcomes17(1), 106 (2019)

    Ayilara, O.F., Zhang, L., Sajobi, T.T., et al.: Impact of missing data on bias and precision when estimating change in patient-reported outcomes from a clinical registry. Health and quality of life outcomes17(1), 106 (2019)

    work page 2019

  4. [4]

    In: Proceedings of the 44th International Con- ference on Distributed Computing Systems

    Cai, S., Zhang, Y., Luo, X., et al.: Robust categorical data clustering guided by multi-granular competitive learning. In: Proceedings of the 44th International Con- ference on Distributed Computing Systems. pp. 288–299. IEEE (2024)

    work page 2024

  5. [5]

    Analog Bits: Generating Discrete Data using Diffusion Models with Self-Conditioning

    Chen, T., Zhang, R., Hinton, G.: Analog bits: Generating discrete data using dif- fusion models with self-conditioning. arXiv preprint arXiv:2208.04202 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  6. [6]

    Advances in Neural Information Processing Systems37, 112050–112103 (2024)

    Chen, Z., Li, H., Wang, F., et al.: Rethinking the diffusion models for missing data imputation: A gradient flow perspective. Advances in Neural Information Processing Systems37, 112050–112103 (2024)

    work page 2024

  7. [7]

    In: Proceedings of the 35th International Conference on Machine Learning

    Cremer, C., Li, X., Duvenaud, D.: Inference suboptimality in variational autoen- coders. In: Proceedings of the 35th International Conference on Machine Learning. pp. 1078–1086. PMLR (2018)

    work page 2018

  8. [8]

    Ad- vances in Neural Information Processing Systems26(2013)

    Cuturi, M.: Sinkhorn distances: Lightspeed computation of optimal transport. Ad- vances in Neural Information Processing Systems26(2013)

    work page 2013

  9. [9]

    Journal ofthe royalstatistical society: seriesB (method- ological)39(1), 1–22 (1977)

    Dempster, A.P., Laird, N.M., Rubin, D.B.: Maximum likelihood from incomplete data viathe emalgorithm. Journal ofthe royalstatistical society: seriesB (method- ological)39(1), 1–22 (1977)

    work page 1977

  10. [10]

    In: Proceedings of the 12th International Conference on Learning Rep- resentations (2024)

    Du, T., Melis, L., Wang, T.: Remasker: Imputing tabular data with masked au- toencoding. In: Proceedings of the 12th International Conference on Learning Rep- resentations (2024)

    work page 2024

  11. [11]

    Advances in neural information processing systems27(2014)

    Goodfellow, I.J., Pouget-Abadie, J., Mirza, M., et al.: Generative adversarial nets. Advances in neural information processing systems27(2014)

    work page 2014

  12. [12]

    Data mining and knowledge discovery2(3), 283–304 (1998)

    Huang, Z.: Extensions to the k-means algorithm for clustering large data sets with categorical values. Data mining and knowledge discovery2(3), 283–304 (1998)

    work page 1998

  13. [13]

    Expert Systems with Applications255, 124507 (2024)

    Hwang, J., Suh, D.: Cc-gain: Clustering and classification-based generative adver- sarial imputation network for missing electricity consumption data imputation. Expert Systems with Applications255, 124507 (2024)

    work page 2024

  14. [14]

    Auto-Encoding Variational Bayes

    Kingma, D.P., Welling, M.: Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114 (2013)

    work page internal anchor Pith review Pith/arXiv arXiv 2013

  15. [15]

    Advances in Neural Information Processing Systems34, 23806–23817 (2021)

    Kyono, T., Zhang, Y., Bellot, A., van der Schaar, M.: Miracle: Causally-aware imputation via learning missing data mechanisms. Advances in Neural Information Processing Systems34, 23806–23817 (2021)

    work page 2021

  16. [16]

    MisGAN: Learning from Incomplete Data with Generative Adversarial Networks

    Li, S.C.X., Jiang, B., Marlin, B.: Misgan: Learning from incomplete data with generative adversarial networks. arXiv preprint arXiv:1902.09599 (2019)

    work page internal anchor Pith review Pith/arXiv arXiv 1902

  17. [17]

    Diffusion and Flow Matching Models for Tabular Data: A Survey

    Li, Z., Huang, Q., Yang, L., et al.: Diffusion models for tabular data: Challenges, current progress, and future directions. arXiv preprint arXiv:2502.17119 (2025)

    work page internal anchor Pith review Pith/arXiv arXiv 2025

  18. [18]

    John Wiley & Sons, 3rd edn

    Little, R.J., Rubin, D.B.: Statistical analysis with missing data. John Wiley & Sons, 3rd edn. (2019)

    work page 2019

  19. [19]

    In: Proceedings of the 33rd ACM International Conference on Information and Knowledge Management

    Liu, Y., Ajanthan, T., Husain, H., Nguyen, V.: Self-supervision improves diffusion models for tabular data imputation. In: Proceedings of the 33rd ACM International Conference on Information and Knowledge Management. pp. 1513–1522 (2024) Cluster-Aware Generative Imputation 17

    work page 2024

  20. [20]

    In: Proceedings of the 36th International Conference on Ma- chine Learning

    Mattei, P.A., Frellsen, J.: Miwae: Deep generative modelling and imputation of incomplete data sets. In: Proceedings of the 36th International Conference on Ma- chine Learning. pp. 4413–4423. PMLR (2019)

    work page 2019

  21. [21]

    In: Proceedings of the 37th International Conference on Machine Learning

    Muzellec, B., Josse, J., Boyer, C., Cuturi, M.: Missing data imputation using op- timal transport. In: Proceedings of the 37th International Conference on Machine Learning. pp. 7130–7140. PMLR (2020)

    work page 2020

  22. [22]

    Pattern Recognition107, 107501 (2020)

    Nazabal, A., Olmos, P.M., Ghahramani, Z., Valera, I.: Handling incomplete het- erogeneous data using vaes. Pattern Recognition107, 107501 (2020)

    work page 2020

  23. [23]

    In: Proceedings of the 3rd International Conference on Contemporary Computing

    Patil, B.M., Joshi, R.C., Toshniwal, D.: Missing value imputation based on k- mean clustering with weighted distance. In: Proceedings of the 3rd International Conference on Contemporary Computing. pp. 600–609. Springer (2010)

    work page 2010

  24. [24]

    Foundations and Trends®in Machine Learning11(5-6), 355–607 (2019)

    Peyré, G., Cuturi, M.: Computational optimal transport: With applications to data science. Foundations and Trends®in Machine Learning11(5-6), 355–607 (2019)

    work page 2019

  25. [25]

    Soft Computing24(6), 4361–4392 (2020)

    Raja, P., Thangavel, K.: Missing value imputation using unsupervised machine learning techniques. Soft Computing24(6), 4361–4392 (2020)

    work page 2020

  26. [26]

    Advances in neural information processing systems29(2016)

    Salimans, T., Goodfellow, I., Zaremba, W., et al.: Improved techniques for training gans. Advances in neural information processing systems29(2016)

    work page 2016

  27. [27]

    Knowledge-Based Systems 249, 108968 (2022)

    Samad,M.D.,Abrar,S.,Diawara,N.:Missingvalueestimationusingclusteringand deep learning within multiple imputation framework. Knowledge-Based Systems 249, 108968 (2022)

    work page 2022

  28. [28]

    Denoising Diffusion Implicit Models

    Song, J., Meng, C., Ermon, S.: Denoising diffusion implicit models. arXiv preprint arXiv:2010.02502 (2020)

    work page internal anchor Pith review Pith/arXiv arXiv 2010

  29. [29]

    Bioinformatics28(1), 112–118 (2012)

    Stekhoven, D.J., Bühlmann, P.: Missforest—non-parametric missing value impu- tation for mixed-type data. Bioinformatics28(1), 112–118 (2012)

    work page 2012

  30. [30]

    In: Proceedings of the 2018 Inter- national Conference on Artificial Intelligence and Statistics

    Tomczak, J., Welling, M.: Vae with a vampprior. In: Proceedings of the 2018 Inter- national Conference on Artificial Intelligence and Statistics. pp. 1214–1223. PMLR (2018)

    work page 2018

  31. [31]

    Bioinformatics17(6), 520–525 (2001)

    Troyanskaya, O., Cantor, M., Sherlock, G.a.: Missing value estimation methods for dna microarrays. Bioinformatics17(6), 520–525 (2001)

    work page 2001

  32. [32]

    Journal of statistical software45, 1–67 (2011)

    Van Buuren, S., Groothuis-Oudshoorn, K.: mice: Multivariate imputation by chained equations in r. Journal of statistical software45, 1–67 (2011)

    work page 2011

  33. [33]

    In: Proceedings of the 35th International Conference on Machine Learning

    Yoon, J., Jordon, J., Schaar, M.: Gain: Missing data imputation using generative adversarial nets. In: Proceedings of the 35th International Conference on Machine Learning. pp. 5689–5698. PMLR (2018)

    work page 2018

  34. [34]

    Advances in Neural Information Processing Systems33, 19075–19087 (2020)

    You, J., Ma, X., Ding, Y., et al.: Handling missing data with graph representation learning. Advances in Neural Information Processing Systems33, 19075–19087 (2020)

    work page 2020

  35. [35]

    In: Proceedings of the the 13th International Conference on Learning Representations (2025)

    Zhang, H., Fang, L., Wu, Q., Yu, P.S.: Diffputer: Empowering diffusion models for missing data imputation. In: Proceedings of the the 13th International Conference on Learning Representations (2025)

    work page 2025

  36. [36]

    In: Transactions on computational science I, pp

    Zhang, S., Zhang, J., Zhu, X., et al.: Missing value imputation based on data clus- tering. In: Transactions on computational science I, pp. 128–138. Springer (2008)

    work page 2008

  37. [37]

    In: Proceedings of the 34th AAAI Conference on Artificial In- telligence

    Zhang, Y., Cheung, Y.M.: An ordinal data clustering algorithm with automated distance learning. In: Proceedings of the 34th AAAI Conference on Artificial In- telligence. vol. 34, pp. 6869–6876 (2020)

    work page 2020

  38. [38]

    IEEE Transactions on Neural Networks and Learning Systems34(9), 6530–6544 (2022)

    Zhang, Y., Cheung, Y.M.: Graph-based dissimilarity measurement for cluster anal- ysis of any-type-attributed data. IEEE Transactions on Neural Networks and Learning Systems34(9), 6530–6544 (2022)

    work page 2022

  39. [39]

    Zhang et al

    Zhang, Y., Cheung, Y.M., Zeng, A.: Het2hom: Representation of heterogeneous at- tributes into homogeneous concept spaces for categorical-and-numerical-attribute 18 C. Zhang et al. data clustering. In: Proceedings of the 31st International Joint Conference on Ar- tificial Intelligence. pp. 3758–3765 (2022)

    work page 2022

  40. [40]

    Proceedings of the ACM on Management of Data3(6), 1–24 (2025)

    Zhang, Y., Zhao, M., Jia, H., et al.: Categorical data clustering via value order estimated distance metric learning. Proceedings of the ACM on Management of Data3(6), 1–24 (2025)

    work page 2025

  41. [41]

    Complex & Intelligent Systems11(1), 84 (2025)

    Zhang, Y., Zou, R., Zhang, Y., et al.: Adaptive micro partition and hierarchical merging for accurate mixed data clustering. Complex & Intelligent Systems11(1), 84 (2025)

    work page 2025

  42. [42]

    In: Proceedings of the 26th European Confer- ence on Artificial Intelligence

    Zhang, Z., Zhang, Y., Zeng, A., et al.: Time-series data imputation via realistic masking-guided tri-attention bi-gru. In: Proceedings of the 26th European Confer- ence on Artificial Intelligence. pp. 3074–3082 (2023)

    work page 2023

  43. [43]

    In: Proceedings of the 40th International Conference on Machine Learning

    Zhao, H., Sun, K., Dezfouli, A., Bonilla, E.V.: Transformed distribution matching for missing value imputation. In: Proceedings of the 40th International Conference on Machine Learning. pp. 42159–42186. PMLR (2023)

    work page 2023

  44. [44]

    In: Proceedings of the 9th International Conference on Data Science and Advanced Analytics

    Zhao, L., Zhang, Y., Ji, Y., et al.: Heterogeneous drift learning: classification of mix-attribute data with concept drifts. In: Proceedings of the 9th International Conference on Data Science and Advanced Analytics. pp. 1–10. IEEE (2022)

    work page 2022

  45. [45]

    In: Proceedings of the 27th European Conference on Artificial Intelligence, pp

    Zhao, M., Feng, S., Zhang, Y., et al.: Learning order forest for qualitative-attribute data clustering. In: Proceedings of the 27th European Conference on Artificial Intelligence, pp. 1943–1950. IOS Press (2024)

    work page 1943

  46. [46]

    In: Proceedings of the 40th AAAI Conference on Artificial Intelligence

    Zhao, M., Huang, Z., Lu, Y., et al.: Break the tie: Learning cluster-customized category relationships for categorical data clustering. In: Proceedings of the 40th AAAI Conference on Artificial Intelligence. vol. 40, pp. 28715–28723 (2026)

    work page 2026

  47. [47]

    In: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining

    Zhao, Y., Udell, M.: Missing value imputation for mixed data via gaussian copula. In: Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. pp. 636–646 (2020)

    work page 2020

  48. [48]

    Diffusion models for missing value imputation in tabular data

    Zheng, S., Charoenphakdee, N.: Diffusion models for missing value imputation in tabular data. arXiv preprint arXiv:2210.17128 (2022)

    work page internal anchor Pith review Pith/arXiv arXiv 2022

  49. [49]

    In: Proceedings of the 37th AAAI Conference on Artificial Intelligence

    Zhong, J., Gui, N., Ye, W.: Data imputation with iterative graph reconstruction. In: Proceedings of the 37th AAAI Conference on Artificial Intelligence. vol. 37, pp. 11399–11407 (2023)

    work page 2023

  50. [50]

    A Survey on Efficient Training of Transformers

    Zhuang, B., Liu, J., Pan, Z., et al.: A survey on efficient training of transformers. arXiv preprint arXiv:2302.01107 (2023)

    work page internal anchor Pith review Pith/arXiv arXiv 2023